def test_wrong_input_shapes():
m = GeometricMeanRelativeAbsoluteError()
with pytest.raises(
ValueError,
match=r"Input data shapes should be the same, but given"):
m.update((torch.rand(4, 1, 2), torch.rand(4, 1)))
with pytest.raises(
ValueError,
match=r"Input data shapes should be the same, but given"):
m.update((torch.rand(4, 1), torch.rand(4, 1, 2)))
with pytest.raises(
ValueError,
match=r"Input data shapes should be the same, but given"):
m.update((
torch.rand(4, 1, 2),
torch.rand(4, ),
))
with pytest.raises(
ValueError,
match=r"Input data shapes should be the same, but given"):
m.update((
torch.rand(4, ),
torch.rand(4, 1, 2),
))
def test_integration():
y_pred = torch.rand(size=(100,))
y = torch.rand(size=(100,))
batch_size = 10
def update_fn(engine, batch):
idx = (engine.state.iteration - 1) * batch_size
y_true_batch = np_y[idx : idx + batch_size]
y_pred_batch = np_y_pred[idx : idx + batch_size]
return torch.from_numpy(y_pred_batch), torch.from_numpy(y_true_batch)
engine = Engine(update_fn)
m = GeometricMeanRelativeAbsoluteError()
m.attach(engine, "gmrae")
np_y = y.numpy().ravel()
np_y_pred = y_pred.numpy().ravel()
data = list(range(y_pred.shape[0] // batch_size))
gmrae = engine.run(data, max_epochs=1).metrics["gmrae"]
sum_errors = np.log(np.abs(np_y - np_y_pred) / np.abs(np_y - np_y.mean())).sum()
np_len = len(y_pred)
np_ans = np.exp(sum_errors / np_len)
assert np_ans == pytest.approx(gmrae)
def test_zero_sample():
m = GeometricMeanRelativeAbsoluteError()
with pytest.raises(
NotComputableError,
match=r"GeometricMeanRelativeAbsoluteError must have at least one example before it can be computed",
):
m.compute()
def _test(n_epochs, metric_device):
metric_device = torch.device(metric_device)
n_iters = 80
s = 16
n_classes = 2
offset = n_iters * s
y_true = torch.rand(size=(offset * idist.get_world_size(),)).to(device)
y_preds = torch.rand(size=(offset * idist.get_world_size(),)).to(device)
def update(engine, i):
return (
y_preds[i * s + rank * offset : (i + 1) * s + rank * offset],
y_true[i * s + rank * offset : (i + 1) * s + rank * offset],
)
engine = Engine(update)
gmrae = GeometricMeanRelativeAbsoluteError(device=metric_device)
gmrae.attach(engine, "gmrae")
data = list(range(n_iters))
engine.run(data=data, max_epochs=n_epochs)
assert "gmrae" in engine.state.metrics
res = engine.state.metrics["gmrae"]
np_y = y_true.cpu().numpy()
np_y_pred = y_preds.cpu().numpy()
np_gmrae = np.exp(np.log(np.abs(np_y - np_y_pred) / np.abs(np_y - np_y.mean())).mean())
assert pytest.approx(res, rel=1e-4) == np_gmrae
def test_wrong_input_shapes():
m = GeometricMeanRelativeAbsoluteError()
with pytest.raises(ValueError):
m.update((torch.rand(4, 1, 2), torch.rand(4, 1)))
with pytest.raises(ValueError):
m.update((torch.rand(4, 1), torch.rand(4, 1, 2)))
with pytest.raises(ValueError):
m.update((torch.rand(4, 1, 2), torch.rand(4, )))
with pytest.raises(ValueError):
m.update((torch.rand(4, ), torch.rand(4, 1, 2)))
def test_integration_geometric_mean_relative_absolute_error_with_output_transform(
):
np.random.seed(1)
size = 105
np_y_pred = np.random.rand(size, 1)
np_y = np.random.rand(size, 1)
np.random.shuffle(np_y)
np_y_sum = 0
num_examples = 0
num_sum_of_errors = 0
np_gmrae = 0
n_iters = 15
batch_size = size // n_iters
for i in range(n_iters + 1):
idx = i * batch_size
np_y_i = np_y[idx:idx + batch_size]
np_y_pred_i = np_y_pred[idx:idx + batch_size]
np_y_sum += np_y_i.sum()
num_examples += np_y_i.shape[0]
np_mean = np_y_sum / num_examples
np_gmrae += np.log(
np.abs(np_y_i - np_y_pred_i) / np.abs(np_y_i - np_mean)).sum()
def update_fn(engine, batch):
idx = (engine.state.iteration - 1) * batch_size
y_true_batch = np_y[idx:idx + batch_size]
y_pred_batch = np_y_pred[idx:idx + batch_size]
return idx, torch.from_numpy(y_pred_batch), torch.from_numpy(
y_true_batch)
engine = Engine(update_fn)
m = GeometricMeanRelativeAbsoluteError(
output_transform=lambda x: (x[1], x[2]))
m.attach(engine, "geometric_mean_relative_absolute_error")
data = list(range(size // batch_size))
gmrae = engine.run(
data, max_epochs=1).metrics["geometric_mean_relative_absolute_error"]
assert np.exp(np_gmrae / num_examples) == pytest.approx(m.compute())
def test_geometric_mean_relative_absolute_error_2():
np.random.seed(1)
size = 105
np_y_pred = np.random.rand(size, 1)
np_y = np.random.rand(size, 1)
np.random.shuffle(np_y)
np_y_sum = 0
num_examples = 0
num_sum_of_errors = 0
np_gmrae = 0
m = GeometricMeanRelativeAbsoluteError()
y_pred = torch.from_numpy(np_y_pred)
y = torch.from_numpy(np_y)
m.reset()
n_iters = 15
batch_size = size // n_iters
for i in range(n_iters + 1):
idx = i * batch_size
np_y_i = np_y[idx:idx + batch_size]
np_y_pred_i = np_y_pred[idx:idx + batch_size]
np_y_sum += np_y_i.sum()
num_examples += np_y_i.shape[0]
np_mean = np_y_sum / num_examples
np_gmrae += np.log(
np.abs(np_y_i - np_y_pred_i) / np.abs(np_y_i - np_mean)).sum()
m.update((y_pred[idx:idx + batch_size], y[idx:idx + batch_size]))
assert np.exp(np_gmrae / num_examples) == pytest.approx(m.compute())
def _test(metric_device):
metric_device = torch.device(metric_device)
m = GeometricMeanRelativeAbsoluteError(device=metric_device)
torch.manual_seed(10 + rank)
y_pred = torch.rand(size=(100,), device=device)
y = torch.rand(size=(100,), device=device)
m.update((y_pred, y))
y_pred = idist.all_gather(y_pred)
y = idist.all_gather(y)
np_y = y.cpu().numpy()
np_y_pred = y_pred.cpu().numpy()
np_gmrae = np.exp(np.log(np.abs(np_y - np_y_pred) / np.abs(np_y - np_y.mean())).mean())
assert m.compute() == pytest.approx(np_gmrae, rel=1e-4)
def test_compute():
size = 51
np_y_pred = np.random.rand(size,)
np_y = np.random.rand(size,)
np_gmrae = np.exp(np.log(np.abs(np_y - np_y_pred) / np.abs(np_y - np_y.mean())).mean())
m = GeometricMeanRelativeAbsoluteError()
y_pred = torch.from_numpy(np_y_pred)
y = torch.from_numpy(np_y)
m.reset()
m.update((y_pred, y))
assert np_gmrae == pytest.approx(m.compute())
